Corn cash price forecasting with neural networks

We explore the forecasting issue in a data set of daily corn cash prices from nearly 500 markets across sixteen states: North Dakota, Iowa, Minnesota, Illinois, Indiana, Ohio, Michigan, Missouri, Nebraska, Arkansas, Kentucky, Wisconsin, South Dakota, Kansas, Oklahoma, and Pennsylvania. We focus on univariate neural network (NN) modeling and bivariate NN modeling with futures prices incorporated. Using simple NNs with twenty hidden neurons and two delays leads to forecasting of high accuracy for the one-day ahead horizon. Including futures prices in the models benefits cash price forecasting. These findings are robust to data splitting ratios for model training, validation, and testing, and to different algorithms employed for model estimates. The forecasting framework here is relatively easy to deploy and has potential to be generalized to other commodities. This study contributes to short-term forecasting users? information needs in decision making processes.

Automated summary

This study explores forecasting of daily corn cash prices from multiple states using neural network models, finding that incorporating futures prices improves accuracy and the models are robust and have potential for generalizability.